显著性偏振参量深度稀疏特征学习的目标检测方法

Object Detection by Deep Sparse Feature Learning of Salient Polarization Parameters

基于偏振成像特点和深层特征分类需求,提出一种显著性偏振参量深度稀疏特征学习的目标检测方法。首先在偏振解析基础上构造显著性偏振参量图像作为检测源图像;然后在判别式字典对下对待检测图像进行稀疏特征学习,并以字典对作为分类器在卷积神经网络(CNN)框架下进行目标分类和定位;最后结合偏振成像探测实际应用需求,选择典型目标和应用场景进行数据采集和模型训练,并进行仿真验证。结果表明该方法在检测得分和平均检测精度上都比直接偏振方向方法有所提高,验证了其有效性,该方法对于有效提升偏振成像探测能力具有应用价值。

Based on polarization imaging characteristics and deep feature classification requirements,an object detection method based on deep sparse feature learning of salient polarization parameters is proposed.First,the salient polarization parameter image is constructed as the source image based on polarization analysis.Then the sparse feature of the image to be detected is learned by discriminant dictionary pair,and the object is classified and located by the dictionary pair which is used as the classifier in CNN framework.Finally,the typical object and scene are selected for data acquisition and model training according to the practical application requirements of polarization imaging detection,and some simulation experiments are conducted.The results show that the detecting score and average detection precision of the proposed method are improved at different degrees by comparing to the polarization direction detection methods and the effectiveness of this method is verified.The proposed method has application value for improving the detection ability of polarization imaging effectively.